So many species, so little time

Tamar Keasar and Elad Chiel - Department of Biology and the Environment – Oranim

 Artificial Intelligence


Seed Grant 2021

Natural history collections document the diversity of life around the world, and allow us to track how it changes over time. They are a key resource for studying the distribution of animals, plants and microbes that were previously unknown, have gone extinct, or have become invasive. However, studying specimens in collections is difficult and time consuming, especially when it comes to many tiny specimens. As a result, new samples accumulate in collections much faster than they are analyzed. Worldwide, more than twenty years may elapse between the collection of a new species from nature and its identification by an expert.

To reduce this bottle-neck in biodiversity research, we are developing methods to help expert biologists find “promising” samples in collections. These may be, for example, samples that were collected from new locations, new capture methods, or over different seasons. Work on such samples should be prioritized as they are the most likely to contain important biological information, such as new species.

In collaboration with colleagues from Israel and the USA (Igor Kleiner, Efrat Gavish-Regev, Michal Segoli, Miriam Kishinevsky, Yehuda Agus), we first curated and explored a large insect collection database. We compared several strategies for selecting samples from this collection for analysis. For example, we tested whether samples collected in the spring are more promising than those collected in the autumn. We also checked whether capturing insects with nets or by vacuuming (see photos) generates equally promising samples. We found that features such as place or season of collection predict how many new species are likely to be found in a sample. We then developed a software that learns these predictors, and generates recommendations for the order of collection samples to be analyzed. This program is similar in spirit to the recommendation systems used by recreation websites (such as Netflix or Youtube) to help users choose music or films. Our software doubles the efficiency of detecting new species in a collection, compared to random sampling. We plan to implement it into a user-friendly web interface, providing a novel research tool for natural history collections worldwide. 



Collecting insects during field research. Top: using nets. Bottom: by vacuuming. Photos: M. Kishinevsky.

Insect samples from field research, to be analyzed in a natural history collection.


Photo: M. Kishinevsky.